Introduction

TODO: Fill in

TAZ Boundary Updates

The zonal system for TRMG2 was inherited from the existing model, and Caliper performed a brief review and recommended additional zone splits. These recommendations were focused on addressing two concerns:

  1. Breaking up zones with large employment and households (in base or future).
  2. Adding granularity for universities to improve modeling of non-auto modes.

Large zones can create traffic loading problems by generating lots of trips and loading them on the network in a small number of locations. Breaking these into smaller zones allows for more-even loading of traffic and avoids spikes in congestion.

A major concern for universities at the outset is getting transit loading correct. Larger zones capture more trips as “intra-zonal” meaning they never leave the zone and load onto the network. For universities, where trips between classes have a significant impact on transit ridership, it is important that these trips travel between zones. In addition, the added granularity gives more accurate measurements of non-motorized travel times, which is important for the university market.

Caliper made several presentations of proposed changes and took feedback from the stakeholders concerning zone edits. The map below shows the final edits. Scroll/Zoom the map below to review them, and use the layers control box to turn the old and new zone layers off.

Existing and Split Zones

Network updates

For each new TAZ, a corresponding centroid node and appropriate centroid connectors were added to the roadway network.

Allocating existing SE data

Placeholder

For now, the SE data is allocated into the new zones using the percent area.

SE Data Updates

To improve model performance, Caliper added several new variables to the SE data:

  • Median household income
  • Workers
  • Age (percent children and seniors)
  • Percent high earning jobs (>$40k)
  • Off-campus students

Median income, workers, and age will improve the synthetic population and allow for better disaggregate trip production models. For instance, being a senior influences trip making behavior.

Understanding which jobs are high earning will help address destination choice issues - particularly Research Triangle Park. Finally, knowing off-campus student locations will improve the university models.

Each new variable is discussed in more detail below.

Household Income

Due to the skewed nature of income distributions, median income is a better approximation of the average than the mean. This information was appended to the socio-economic (SE) data from the Census, but income information is only available at the block group geography level. As a result, all TAZs within a block group have the same median income. In addition, income measures were suppressed by the Census for 88 block groups. For TAZs in these block groups, tract median income was used. If the tract data was also missing, then the county median was used (shown below). This information is stored in the Median_Inc field of the SE table.

County Median Income
Alamance $45,735
Chatham $63,531
Durham $58,190
Franklin $53,175
Granville $55,628
Harnett $51,630
Johnston $56,842
Nash $48,362
Orange $68,211
Person $48,811
Wake $76,956

Workers

Estimates of workers are only available from the ACS at the tract level. It is also important to note that ACS tables that count total workers cannot be used, because they include workers living in group quarters. The model uses this worker information to build a synthetic population living in households, and including workers in group quarters would bias the results. Instead, total workers in households must be imputed from tables that count households by number of workers. In other words, each 1-worker household adds 1 worker to the total number of workers. 2-worker households add 2. Each household with 3-or-more workers adds 3.1 workers. This 3.1 metric is not based on a solid data source, and may be refined if one can be found.

After determining the total number of workers in each tract, Caliper:

  1. Calculated the percentage of workers for the ACS tract (HH workers / HH population).
  2. Applied that percentage to each TAZ within the tract.

This information is contained in the Pct_Worker field of the SE data table.

Age

Like income, age data is available from the ACS at the block group level. For each block group, Caliper computed the percentage of children and seniors.

  • Child: Under 18
  • Senior: 65 or older

The ACSs data on age is provided for total population and not population in households. Just as with workers, including group quarters would bias the synthetic household population generated by the model.

Caliper did an extensive review of the ACS data and found that the marginal columns in other tables could be used to find household population totals by the three largest categories:

  • B09019_002: Field 2 provides total household population.
  • B09018_001: Field 1 provides a total count of children in households.
  • B09020_002: Field 2 provides a total count of seniors in households.

The histogram below shows the count of block groups by percent children and seniors. A minimum percentage was asserted for both children and seniors at 2.5%. This prevents TAZs with 1,000 or more household population from having unrealistically low levels of children and seniors (the ACS is a survey, not a full enumeration).

Next, each TAZ within a block group was assigned the same percentages. Age information was available for all block groups with population, and no TAZs needed to use a tract or county data. This information can be found in the SE data table in the fields Pct_Child and Pct_Senior.

Percent high earnings

The LEHD Origin-Destination Employment Statistics (LODES) provides many valuable attributes including a breakdown of jobs by earnings group at the block level. The trade off for this spatial accuracy is that LODES only provides three income groups:

  • Under $1,250 / month ($15k per year)
  • Between $1,250 and $3,333 per month ($15k and $40k per year)
  • Above $3,333 per month

Due to these limited break points, Caliper split high and low income groups at the $40k annual cut off. The block data was then aggregated to TAZs. For TAZs with less than 30 jobs, the aggregate tract percentage was used instead.

While the $40k earning figure is lower than the median individual income of the region, a map of the percentage of jobs in each category still shows some important patterns. In the map below, note that RTP shows up with a higher percentage of high-earning jobs compared to the zones around it. This helps match high-income households in west Cary with high-earning jobs in RTP.

This information was added to the SE data table in the field PctHighEarn.

Student housing (off-campus)

University travel is an important market for the Triangle, which is home to three major universities and many medium and small colleges. The TRMG2 model takes a different approach to university students than the previous model. For the four largest universities off-campus addresses were collected and are used directly (after some cleaning). Importantly, the students in each university are stored separately in the SE data, which greatly simplifies the destination choice problem.

Total enrollment statistics were collected for the four major universities:

  • North Carolina State University (source)
  • University of North Carolina at Chapel Hill (source)
  • Duke University (source)
  • North Carolina Central University (source)

TRMG2 stakeholders provided counts of students living in dorms for each campus, which allowed Caliper to calculate the total number of off-campus students as shown in the table below.

School Total Enrollment In Dorms Implied Off-Campus
NCSU 31,008 12,424 18,584
UNC 29,468 11,390 18,078
Duke 15,928 5,788 10,140
NCCU 8,096 2,899 5,197

Stakeholders also provided Caliper with off-campus student addresses from the same four universities. The NCSU, UNC, and NCCU data sets contained student addresses from all over the country (and world). These addresses are a mix of local students living off campus, distance ed students, and billing addresses (often parent addresses) as shown in the map below.

The histograms below show the number of address points for each university in one-mile bands around the campus. The intensity drops off drastically beyond 10 miles.

For each school, Caliper ignored address points outside 10 miles as shown below.

Finally, the weight of each point within the buffer was factored up to match the total off-campus enrollment for each university. As an example, after weighting, each point might represent 1.3 students. These weights were then aggregated by TAZ and added to the the off-campus student fields in the SE data. This information is contained in the SE data table in fields StudOff_NCSU, StudOff_UNC, StudOff_Duke, and StudOff_NCCU.





Caliper Corporation, 2021